blob: eae088f6aaaeb6e8a3380adad19fb66feebe9905 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/fs/super.c
3 *
4 * Copyright (C) 1991, 1992 Linus Torvalds
5 *
6 * super.c contains code to handle: - mount structures
7 * - super-block tables
8 * - filesystem drivers list
9 * - mount system call
10 * - umount system call
11 * - ustat system call
12 *
13 * GK 2/5/95 - Changed to support mounting the root fs via NFS
14 *
15 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall
16 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96
17 * Added options to /proc/mounts:
Jan Engelhardt96de0e22007-10-19 23:21:04 +020018 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
Linus Torvalds1da177e2005-04-16 15:20:36 -070019 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
20 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
21 */
22
Paul Gortmaker630d9c42011-11-16 23:57:37 -050023#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/slab.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/blkdev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/mount.h>
27#include <linux/security.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/writeback.h> /* for the emergency remount stuff */
29#include <linux/idr.h>
Ingo Molnar353ab6e2006-03-26 01:37:12 -080030#include <linux/mutex.h>
Jens Axboe5477d0f2010-04-29 20:33:35 +020031#include <linux/backing-dev.h>
Nick Pigginceb5bdc2011-01-07 17:50:05 +110032#include <linux/rculist_bl.h>
Dan Magenheimerc515e1f2011-05-26 10:01:43 -060033#include <linux/cleancache.h>
Al Viro40401532012-02-13 03:58:52 +000034#include <linux/fsnotify.h>
Jan Kara5accdf82012-06-12 16:20:34 +020035#include <linux/lockdep.h>
Al Viro6d59e7f2008-03-22 15:48:17 -040036#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039LIST_HEAD(super_blocks);
40DEFINE_SPINLOCK(sb_lock);
41
Jan Kara5accdf82012-06-12 16:20:34 +020042static char *sb_writers_name[SB_FREEZE_LEVELS] = {
43 "sb_writers",
44 "sb_pagefaults",
45 "sb_internal",
46};
47
Dave Chinnerb0d40c92011-07-08 14:14:42 +100048/*
49 * One thing we have to be careful of with a per-sb shrinker is that we don't
50 * drop the last active reference to the superblock from within the shrinker.
51 * If that happens we could trigger unregistering the shrinker from within the
52 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
53 * take a passive reference to the superblock to avoid this from occurring.
54 */
Dave Chinner0a234c62013-08-28 10:17:57 +100055static unsigned long super_cache_scan(struct shrinker *shrink,
56 struct shrink_control *sc)
Dave Chinnerb0d40c92011-07-08 14:14:42 +100057{
58 struct super_block *sb;
Dave Chinner0a234c62013-08-28 10:17:57 +100059 long fs_objects = 0;
60 long total_objects;
61 long freed = 0;
62 long dentries;
63 long inodes;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100064
65 sb = container_of(shrink, struct super_block, s_shrink);
66
67 /*
68 * Deadlock avoidance. We may hold various FS locks, and we don't want
69 * to recurse into the FS that called us in clear_inode() and friends..
70 */
Dave Chinner0a234c62013-08-28 10:17:57 +100071 if (!(sc->gfp_mask & __GFP_FS))
72 return SHRINK_STOP;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100073
74 if (!grab_super_passive(sb))
Dave Chinner0a234c62013-08-28 10:17:57 +100075 return SHRINK_STOP;
Dave Chinnerb0d40c92011-07-08 14:14:42 +100076
Al Virod0407902013-07-19 21:12:31 +040077 if (sb->s_op->nr_cached_objects)
Dave Chinner9b17c622013-08-28 10:18:05 +100078 fs_objects = sb->s_op->nr_cached_objects(sb, sc->nid);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100079
Dave Chinner9b17c622013-08-28 10:18:05 +100080 inodes = list_lru_count_node(&sb->s_inode_lru, sc->nid);
81 dentries = list_lru_count_node(&sb->s_dentry_lru, sc->nid);
Dave Chinnerf6041562013-08-28 10:18:00 +100082 total_objects = dentries + inodes + fs_objects + 1;
Tetsuo Handa475d0db2014-05-17 20:56:38 +090083 if (!total_objects)
84 total_objects = 1;
Dave Chinner0e1fdaf2011-07-08 14:14:44 +100085
Dave Chinner0a234c62013-08-28 10:17:57 +100086 /* proportion the scan between the caches */
Dave Chinnerf6041562013-08-28 10:18:00 +100087 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects);
Dave Chinnerbc3b14c2013-08-28 10:17:58 +100088 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects);
Dave Chinnerb0d40c92011-07-08 14:14:42 +100089
Dave Chinner0a234c62013-08-28 10:17:57 +100090 /*
91 * prune the dcache first as the icache is pinned by it, then
92 * prune the icache, followed by the filesystem specific caches
93 */
Dave Chinner9b17c622013-08-28 10:18:05 +100094 freed = prune_dcache_sb(sb, dentries, sc->nid);
95 freed += prune_icache_sb(sb, inodes, sc->nid);
Dave Chinnerb0d40c92011-07-08 14:14:42 +100096
Dave Chinner0a234c62013-08-28 10:17:57 +100097 if (fs_objects) {
98 fs_objects = mult_frac(sc->nr_to_scan, fs_objects,
99 total_objects);
Dave Chinner9b17c622013-08-28 10:18:05 +1000100 freed += sb->s_op->free_cached_objects(sb, fs_objects,
101 sc->nid);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000102 }
103
Dave Chinner0a234c62013-08-28 10:17:57 +1000104 drop_super(sb);
105 return freed;
106}
107
108static unsigned long super_cache_count(struct shrinker *shrink,
109 struct shrink_control *sc)
110{
111 struct super_block *sb;
112 long total_objects = 0;
113
114 sb = container_of(shrink, struct super_block, s_shrink);
115
Tim Chend23da152014-06-04 16:10:47 -0700116 /*
117 * Don't call grab_super_passive as it is a potential
118 * scalability bottleneck. The counts could get updated
119 * between super_cache_count and super_cache_scan anyway.
120 * Call to super_cache_count with shrinker_rwsem held
121 * ensures the safety of call to list_lru_count_node() and
122 * s_op->nr_cached_objects().
123 */
Dave Chinner0a234c62013-08-28 10:17:57 +1000124 if (sb->s_op && sb->s_op->nr_cached_objects)
Dave Chinner9b17c622013-08-28 10:18:05 +1000125 total_objects = sb->s_op->nr_cached_objects(sb,
126 sc->nid);
Dave Chinner0a234c62013-08-28 10:17:57 +1000127
Dave Chinner9b17c622013-08-28 10:18:05 +1000128 total_objects += list_lru_count_node(&sb->s_dentry_lru,
129 sc->nid);
130 total_objects += list_lru_count_node(&sb->s_inode_lru,
131 sc->nid);
Dave Chinner0a234c62013-08-28 10:17:57 +1000132
Glauber Costa55f841c2013-08-28 10:17:53 +1000133 total_objects = vfs_pressure_ratio(total_objects);
Dave Chinner0e1fdaf2011-07-08 14:14:44 +1000134 return total_objects;
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000135}
136
Al Viro7eb5e882013-10-01 15:09:58 -0400137/**
138 * destroy_super - frees a superblock
139 * @s: superblock to free
140 *
141 * Frees a superblock.
142 */
143static void destroy_super(struct super_block *s)
Jan Kara5accdf82012-06-12 16:20:34 +0200144{
145 int i;
Al Viro7eb5e882013-10-01 15:09:58 -0400146 list_lru_destroy(&s->s_dentry_lru);
147 list_lru_destroy(&s->s_inode_lru);
Jan Kara5accdf82012-06-12 16:20:34 +0200148 for (i = 0; i < SB_FREEZE_LEVELS; i++)
149 percpu_counter_destroy(&s->s_writers.counter[i]);
Al Viro7eb5e882013-10-01 15:09:58 -0400150 security_sb_free(s);
151 WARN_ON(!list_empty(&s->s_mounts));
152 kfree(s->s_subtype);
153 kfree(s->s_options);
Al Viroe2fec7c2013-10-04 17:06:56 -0400154 kfree_rcu(s, rcu);
Jan Kara5accdf82012-06-12 16:20:34 +0200155}
156
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157/**
158 * alloc_super - create new superblock
Henrik Kretzschmarfe2bbc42006-09-06 00:03:41 -0700159 * @type: filesystem type superblock should belong to
David Howells9249e172012-06-25 12:55:37 +0100160 * @flags: the mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 *
162 * Allocates and initializes a new &struct super_block. alloc_super()
163 * returns a pointer new superblock or %NULL if allocation had failed.
164 */
David Howells9249e172012-06-25 12:55:37 +0100165static struct super_block *alloc_super(struct file_system_type *type, int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166{
Oliver Neukum11b0b5a2006-03-25 03:08:13 -0800167 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER);
Alexey Dobriyanb87221d2009-09-21 17:01:09 -0700168 static const struct super_operations default_op;
Al Viro7eb5e882013-10-01 15:09:58 -0400169 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170
Al Viro7eb5e882013-10-01 15:09:58 -0400171 if (!s)
172 return NULL;
173
Vladimir Davydovb5bd8562014-01-21 15:48:45 -0800174 INIT_LIST_HEAD(&s->s_mounts);
175
Al Viro7eb5e882013-10-01 15:09:58 -0400176 if (security_sb_alloc(s))
177 goto fail;
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200178
Al Viro7eb5e882013-10-01 15:09:58 -0400179 for (i = 0; i < SB_FREEZE_LEVELS; i++) {
Tejun Heo908c7f12014-09-08 09:51:29 +0900180 if (percpu_counter_init(&s->s_writers.counter[i], 0,
181 GFP_KERNEL) < 0)
Al Viro7eb5e882013-10-01 15:09:58 -0400182 goto fail;
183 lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i],
184 &type->s_writers_key[i], 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 }
Al Viro7eb5e882013-10-01 15:09:58 -0400186 init_waitqueue_head(&s->s_writers.wait);
187 init_waitqueue_head(&s->s_writers.wait_unfrozen);
188 s->s_flags = flags;
189 s->s_bdi = &default_backing_dev_info;
190 INIT_HLIST_NODE(&s->s_instances);
191 INIT_HLIST_BL_HEAD(&s->s_anon);
192 INIT_LIST_HEAD(&s->s_inodes);
193
194 if (list_lru_init(&s->s_dentry_lru))
195 goto fail;
196 if (list_lru_init(&s->s_inode_lru))
197 goto fail;
198
Al Viro7eb5e882013-10-01 15:09:58 -0400199 init_rwsem(&s->s_umount);
200 lockdep_set_class(&s->s_umount, &type->s_umount_key);
201 /*
202 * sget() can have s_umount recursion.
203 *
204 * When it cannot find a suitable sb, it allocates a new
205 * one (this one), and tries again to find a suitable old
206 * one.
207 *
208 * In case that succeeds, it will acquire the s_umount
209 * lock of the old one. Since these are clearly distrinct
210 * locks, and this object isn't exposed yet, there's no
211 * risk of deadlocks.
212 *
213 * Annotate this by putting this lock in a different
214 * subclass.
215 */
216 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
217 s->s_count = 1;
218 atomic_set(&s->s_active, 1);
219 mutex_init(&s->s_vfs_rename_mutex);
220 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
221 mutex_init(&s->s_dquot.dqio_mutex);
222 mutex_init(&s->s_dquot.dqonoff_mutex);
Al Viro7eb5e882013-10-01 15:09:58 -0400223 s->s_maxbytes = MAX_NON_LFS;
224 s->s_op = &default_op;
225 s->s_time_gran = 1000000000;
226 s->cleancache_poolid = -1;
227
228 s->s_shrink.seeks = DEFAULT_SEEKS;
229 s->s_shrink.scan_objects = super_cache_scan;
230 s->s_shrink.count_objects = super_cache_count;
231 s->s_shrink.batch = 1024;
232 s->s_shrink.flags = SHRINKER_NUMA_AWARE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233 return s;
Glauber Costa5ca302c2013-08-28 10:18:18 +1000234
Al Viro7eb5e882013-10-01 15:09:58 -0400235fail:
236 destroy_super(s);
237 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700238}
239
240/* Superblock refcounting */
241
242/*
Al Viro35cf7ba2010-03-22 21:13:53 -0400243 * Drop a superblock's refcount. The caller must hold sb_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244 */
Al Virof47ec3f2011-11-21 21:15:42 -0500245static void __put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 if (!--sb->s_count) {
Al Viro551de6f2010-03-22 19:36:35 -0400248 list_del_init(&sb->s_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 destroy_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251}
252
253/**
254 * put_super - drop a temporary reference to superblock
255 * @sb: superblock in question
256 *
257 * Drops a temporary reference, frees superblock if there's no
258 * references left.
259 */
Al Virof47ec3f2011-11-21 21:15:42 -0500260static void put_super(struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261{
262 spin_lock(&sb_lock);
263 __put_super(sb);
264 spin_unlock(&sb_lock);
265}
266
267
268/**
Al Viro74dbbdd2009-05-06 01:07:50 -0400269 * deactivate_locked_super - drop an active reference to superblock
270 * @s: superblock to deactivate
271 *
Al Viro1712ac82010-03-22 15:22:31 -0400272 * Drops an active reference to superblock, converting it into a temprory
273 * one if there is no other active references left. In that case we
274 * tell fs driver to shut it down and drop the temporary reference we
275 * had just acquired.
276 *
277 * Caller holds exclusive lock on superblock; that lock is released.
Al Viro74dbbdd2009-05-06 01:07:50 -0400278 */
279void deactivate_locked_super(struct super_block *s)
280{
281 struct file_system_type *fs = s->s_type;
Al Virob20bd1a2010-03-22 08:53:19 -0400282 if (atomic_dec_and_test(&s->s_active)) {
Dan Magenheimer31677602011-09-21 11:56:28 -0400283 cleancache_invalidate_fs(s);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000284 unregister_shrinker(&s->s_shrink);
Dave Chinner28f2cd4f2014-06-04 16:10:46 -0700285 fs->kill_sb(s);
Glauber Costaf5e1dd32013-08-28 10:18:18 +1000286
Al Viro74dbbdd2009-05-06 01:07:50 -0400287 put_filesystem(fs);
288 put_super(s);
289 } else {
290 up_write(&s->s_umount);
291 }
292}
293
294EXPORT_SYMBOL(deactivate_locked_super);
295
296/**
Al Viro1712ac82010-03-22 15:22:31 -0400297 * deactivate_super - drop an active reference to superblock
298 * @s: superblock to deactivate
299 *
300 * Variant of deactivate_locked_super(), except that superblock is *not*
301 * locked by caller. If we are going to drop the final active reference,
302 * lock will be acquired prior to that.
303 */
304void deactivate_super(struct super_block *s)
305{
306 if (!atomic_add_unless(&s->s_active, -1, 1)) {
307 down_write(&s->s_umount);
308 deactivate_locked_super(s);
309 }
310}
311
312EXPORT_SYMBOL(deactivate_super);
313
314/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 * grab_super - acquire an active reference
316 * @s: reference we are trying to make active
317 *
318 * Tries to acquire an active reference. grab_super() is used when we
319 * had just found a superblock in super_blocks or fs_type->fs_supers
320 * and want to turn it into a full-blown active reference. grab_super()
321 * is called with sb_lock held and drops it. Returns 1 in case of
322 * success, 0 if we had failed (superblock contents was already dead or
Al Viroacfec9a2013-07-20 03:13:55 +0400323 * dying when grab_super() had been called). Note that this is only
324 * called for superblocks not in rundown mode (== ones still on ->fs_supers
325 * of their type), so increment of ->s_count is OK here.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326 */
Josh Triplett9c4dbee2006-09-29 01:59:29 -0700327static int grab_super(struct super_block *s) __releases(sb_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328{
329 s->s_count++;
330 spin_unlock(&sb_lock);
331 down_write(&s->s_umount);
Al Viroacfec9a2013-07-20 03:13:55 +0400332 if ((s->s_flags & MS_BORN) && atomic_inc_not_zero(&s->s_active)) {
333 put_super(s);
334 return 1;
335 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 up_write(&s->s_umount);
337 put_super(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 return 0;
339}
340
David Howellscf9a2ae2006-08-29 19:05:54 +0100341/*
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000342 * grab_super_passive - acquire a passive reference
Wanpeng Li331cbde2012-06-09 11:10:55 +0800343 * @sb: reference we are trying to grab
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000344 *
345 * Tries to acquire a passive reference. This is used in places where we
346 * cannot take an active reference but we need to ensure that the
347 * superblock does not go away while we are working on it. It returns
348 * false if a reference was not gained, and returns true with the s_umount
349 * lock held in read mode if a reference is gained. On successful return,
350 * the caller must drop the s_umount lock and the passive reference when
351 * done.
352 */
353bool grab_super_passive(struct super_block *sb)
354{
355 spin_lock(&sb_lock);
Al Viroa5166162011-12-12 22:53:00 -0500356 if (hlist_unhashed(&sb->s_instances)) {
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000357 spin_unlock(&sb_lock);
358 return false;
359 }
360
361 sb->s_count++;
362 spin_unlock(&sb_lock);
363
364 if (down_read_trylock(&sb->s_umount)) {
Al Virodabe0dc2012-01-03 21:01:29 -0500365 if (sb->s_root && (sb->s_flags & MS_BORN))
Dave Chinner12ad3ab2011-07-08 14:14:41 +1000366 return true;
367 up_read(&sb->s_umount);
368 }
369
370 put_super(sb);
371 return false;
372}
373
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374/**
375 * generic_shutdown_super - common helper for ->kill_sb()
376 * @sb: superblock to kill
377 *
378 * generic_shutdown_super() does all fs-independent work on superblock
379 * shutdown. Typical ->kill_sb() should pick all fs-specific objects
380 * that need destruction out of superblock, call generic_shutdown_super()
381 * and release aforementioned objects. Note: dentries and inodes _are_
382 * taken care of and do not need specific handling.
David Howellsc636ebd2006-10-11 01:22:19 -0700383 *
384 * Upon calling this function, the filesystem may no longer alter or
385 * rearrange the set of dentries belonging to this super_block, nor may it
386 * change the attachments of dentries to inodes.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 */
388void generic_shutdown_super(struct super_block *sb)
389{
Josef 'Jeff' Sipekee9b6d62007-02-12 00:55:41 -0800390 const struct super_operations *sop = sb->s_op;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391
David Howellsc636ebd2006-10-11 01:22:19 -0700392 if (sb->s_root) {
393 shrink_dcache_for_umount(sb);
Jan Kara60b06802009-04-27 16:43:53 +0200394 sync_filesystem(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395 sb->s_flags &= ~MS_ACTIVE;
Arjan van de Venefaee192009-01-06 07:20:54 -0800396
Al Viro63997e92010-10-25 20:49:35 -0400397 fsnotify_unmount_inodes(&sb->s_inodes);
398
399 evict_inodes(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700400
Christoph Hellwig7b7a8662013-09-04 15:04:39 +0200401 if (sb->s_dio_done_wq) {
402 destroy_workqueue(sb->s_dio_done_wq);
403 sb->s_dio_done_wq = NULL;
404 }
405
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 if (sop->put_super)
407 sop->put_super(sb);
408
Al Viro63997e92010-10-25 20:49:35 -0400409 if (!list_empty(&sb->s_inodes)) {
Dave Jones7b4fe292006-02-07 12:58:48 -0800410 printk("VFS: Busy inodes after unmount of %s. "
411 "Self-destruct in 5 seconds. Have a nice day...\n",
412 sb->s_id);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 }
415 spin_lock(&sb_lock);
416 /* should be initialized for __put_super_and_need_restart() */
Al Viroa5166162011-12-12 22:53:00 -0500417 hlist_del_init(&sb->s_instances);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 spin_unlock(&sb_lock);
419 up_write(&sb->s_umount);
420}
421
422EXPORT_SYMBOL(generic_shutdown_super);
423
424/**
425 * sget - find or create a superblock
426 * @type: filesystem type superblock should belong to
427 * @test: comparison callback
428 * @set: setup callback
David Howells9249e172012-06-25 12:55:37 +0100429 * @flags: mount flags
Linus Torvalds1da177e2005-04-16 15:20:36 -0700430 * @data: argument to each of them
431 */
432struct super_block *sget(struct file_system_type *type,
433 int (*test)(struct super_block *,void *),
434 int (*set)(struct super_block *,void *),
David Howells9249e172012-06-25 12:55:37 +0100435 int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 void *data)
437{
438 struct super_block *s = NULL;
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700439 struct super_block *old;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 int err;
441
442retry:
443 spin_lock(&sb_lock);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700444 if (test) {
Sasha Levinb67bfe02013-02-27 17:06:00 -0800445 hlist_for_each_entry(old, &type->fs_supers, s_instances) {
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700446 if (!test(old, data))
447 continue;
448 if (!grab_super(old))
449 goto retry;
Li Zefana3cfbb52009-03-12 14:31:29 -0700450 if (s) {
451 up_write(&s->s_umount);
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700452 destroy_super(s);
Al Viro7a4dec52010-08-09 12:05:43 -0400453 s = NULL;
Li Zefana3cfbb52009-03-12 14:31:29 -0700454 }
Matthias Kaehlcked4730122007-10-18 23:39:57 -0700455 return old;
456 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 }
458 if (!s) {
459 spin_unlock(&sb_lock);
David Howells9249e172012-06-25 12:55:37 +0100460 s = alloc_super(type, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461 if (!s)
462 return ERR_PTR(-ENOMEM);
463 goto retry;
464 }
465
466 err = set(s, data);
467 if (err) {
468 spin_unlock(&sb_lock);
Li Zefana3cfbb52009-03-12 14:31:29 -0700469 up_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700470 destroy_super(s);
471 return ERR_PTR(err);
472 }
473 s->s_type = type;
474 strlcpy(s->s_id, type->name, sizeof(s->s_id));
475 list_add_tail(&s->s_list, &super_blocks);
Al Viroa5166162011-12-12 22:53:00 -0500476 hlist_add_head(&s->s_instances, &type->fs_supers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477 spin_unlock(&sb_lock);
478 get_filesystem(type);
Dave Chinnerb0d40c92011-07-08 14:14:42 +1000479 register_shrinker(&s->s_shrink);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 return s;
481}
482
483EXPORT_SYMBOL(sget);
484
485void drop_super(struct super_block *sb)
486{
487 up_read(&sb->s_umount);
488 put_super(sb);
489}
490
491EXPORT_SYMBOL(drop_super);
492
Christoph Hellwige5004752009-05-05 16:08:56 +0200493/**
Al Viro01a05b32010-03-23 06:06:58 -0400494 * iterate_supers - call function for all active superblocks
495 * @f: function to call
496 * @arg: argument to pass to it
497 *
498 * Scans the superblock list and calls given function, passing it
499 * locked superblock and given argument.
500 */
501void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
502{
Al Virodca33252010-07-25 02:31:46 +0400503 struct super_block *sb, *p = NULL;
Al Viro01a05b32010-03-23 06:06:58 -0400504
505 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400506 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500507 if (hlist_unhashed(&sb->s_instances))
Al Viro01a05b32010-03-23 06:06:58 -0400508 continue;
509 sb->s_count++;
510 spin_unlock(&sb_lock);
511
512 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500513 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro01a05b32010-03-23 06:06:58 -0400514 f(sb, arg);
515 up_read(&sb->s_umount);
516
517 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400518 if (p)
519 __put_super(p);
520 p = sb;
Al Viro01a05b32010-03-23 06:06:58 -0400521 }
Al Virodca33252010-07-25 02:31:46 +0400522 if (p)
523 __put_super(p);
Al Viro01a05b32010-03-23 06:06:58 -0400524 spin_unlock(&sb_lock);
525}
526
527/**
Al Viro43e15cd2011-06-03 20:16:57 -0400528 * iterate_supers_type - call function for superblocks of given type
529 * @type: fs type
530 * @f: function to call
531 * @arg: argument to pass to it
532 *
533 * Scans the superblock list and calls given function, passing it
534 * locked superblock and given argument.
535 */
536void iterate_supers_type(struct file_system_type *type,
537 void (*f)(struct super_block *, void *), void *arg)
538{
539 struct super_block *sb, *p = NULL;
540
541 spin_lock(&sb_lock);
Sasha Levinb67bfe02013-02-27 17:06:00 -0800542 hlist_for_each_entry(sb, &type->fs_supers, s_instances) {
Al Viro43e15cd2011-06-03 20:16:57 -0400543 sb->s_count++;
544 spin_unlock(&sb_lock);
545
546 down_read(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500547 if (sb->s_root && (sb->s_flags & MS_BORN))
Al Viro43e15cd2011-06-03 20:16:57 -0400548 f(sb, arg);
549 up_read(&sb->s_umount);
550
551 spin_lock(&sb_lock);
552 if (p)
553 __put_super(p);
554 p = sb;
555 }
556 if (p)
557 __put_super(p);
558 spin_unlock(&sb_lock);
559}
560
561EXPORT_SYMBOL(iterate_supers_type);
562
563/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 * get_super - get the superblock of a device
565 * @bdev: device to get the superblock for
566 *
567 * Scans the superblock list and finds the superblock of the file system
568 * mounted on the device given. %NULL is returned if no match is found.
569 */
570
Al Virodf40c012010-03-22 20:23:25 -0400571struct super_block *get_super(struct block_device *bdev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700573 struct super_block *sb;
574
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 if (!bdev)
576 return NULL;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700577
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700579rescan:
580 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500581 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400582 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700583 if (sb->s_bdev == bdev) {
584 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700586 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400587 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500588 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700589 return sb;
590 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400591 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700592 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400593 __put_super(sb);
594 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 }
596 }
597 spin_unlock(&sb_lock);
598 return NULL;
599}
600
601EXPORT_SYMBOL(get_super);
Christoph Hellwig45042302009-08-03 23:28:35 +0200602
603/**
Jan Kara6b6dc832012-02-10 11:03:00 +0100604 * get_super_thawed - get thawed superblock of a device
605 * @bdev: device to get the superblock for
606 *
607 * Scans the superblock list and finds the superblock of the file system
608 * mounted on the device. The superblock is returned once it is thawed
609 * (or immediately if it was not frozen). %NULL is returned if no match
610 * is found.
611 */
612struct super_block *get_super_thawed(struct block_device *bdev)
613{
614 while (1) {
615 struct super_block *s = get_super(bdev);
Jan Kara5accdf82012-06-12 16:20:34 +0200616 if (!s || s->s_writers.frozen == SB_UNFROZEN)
Jan Kara6b6dc832012-02-10 11:03:00 +0100617 return s;
618 up_read(&s->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +0200619 wait_event(s->s_writers.wait_unfrozen,
620 s->s_writers.frozen == SB_UNFROZEN);
Jan Kara6b6dc832012-02-10 11:03:00 +0100621 put_super(s);
622 }
623}
624EXPORT_SYMBOL(get_super_thawed);
625
626/**
Christoph Hellwig45042302009-08-03 23:28:35 +0200627 * get_active_super - get an active reference to the superblock of a device
628 * @bdev: device to get the superblock for
629 *
630 * Scans the superblock list and finds the superblock of the file system
631 * mounted on the device given. Returns the superblock with an active
Al Virod3f21472010-03-23 11:11:05 -0400632 * reference or %NULL if none was found.
Christoph Hellwig45042302009-08-03 23:28:35 +0200633 */
634struct super_block *get_active_super(struct block_device *bdev)
635{
636 struct super_block *sb;
637
638 if (!bdev)
639 return NULL;
640
Al Viro14945832010-03-22 20:15:33 -0400641restart:
Christoph Hellwig45042302009-08-03 23:28:35 +0200642 spin_lock(&sb_lock);
643 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500644 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400645 continue;
Al Viro14945832010-03-22 20:15:33 -0400646 if (sb->s_bdev == bdev) {
Al Viroacfec9a2013-07-20 03:13:55 +0400647 if (!grab_super(sb))
Al Viro14945832010-03-22 20:15:33 -0400648 goto restart;
Al Viroacfec9a2013-07-20 03:13:55 +0400649 up_write(&sb->s_umount);
650 return sb;
Al Viro14945832010-03-22 20:15:33 -0400651 }
Christoph Hellwig45042302009-08-03 23:28:35 +0200652 }
653 spin_unlock(&sb_lock);
654 return NULL;
655}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656
Al Virodf40c012010-03-22 20:23:25 -0400657struct super_block *user_get_super(dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658{
Kirill Korotaev618f0632005-06-23 00:09:54 -0700659 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 spin_lock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700662rescan:
663 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500664 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400665 continue;
Kirill Korotaev618f0632005-06-23 00:09:54 -0700666 if (sb->s_dev == dev) {
667 sb->s_count++;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 spin_unlock(&sb_lock);
Kirill Korotaev618f0632005-06-23 00:09:54 -0700669 down_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400670 /* still alive? */
Al Virodabe0dc2012-01-03 21:01:29 -0500671 if (sb->s_root && (sb->s_flags & MS_BORN))
Kirill Korotaev618f0632005-06-23 00:09:54 -0700672 return sb;
673 up_read(&sb->s_umount);
Al Virodf40c012010-03-22 20:23:25 -0400674 /* nope, got unmounted */
Kirill Korotaev618f0632005-06-23 00:09:54 -0700675 spin_lock(&sb_lock);
Al Virodf40c012010-03-22 20:23:25 -0400676 __put_super(sb);
677 goto rescan;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700678 }
679 }
680 spin_unlock(&sb_lock);
681 return NULL;
682}
683
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685 * do_remount_sb - asks filesystem to change mount options.
686 * @sb: superblock in question
687 * @flags: numeric part of options
688 * @data: the rest of options
689 * @force: whether or not to force the change
690 *
691 * Alters the mount options of a mounted file system.
692 */
693int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
694{
695 int retval;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400696 int remount_ro;
Christoph Hellwig45042302009-08-03 23:28:35 +0200697
Jan Kara5accdf82012-06-12 16:20:34 +0200698 if (sb->s_writers.frozen != SB_UNFROZEN)
Christoph Hellwig45042302009-08-03 23:28:35 +0200699 return -EBUSY;
700
David Howells93614012006-09-30 20:45:40 +0200701#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
703 return -EACCES;
David Howells93614012006-09-30 20:45:40 +0200704#endif
Christoph Hellwig45042302009-08-03 23:28:35 +0200705
Nick Piggind208bbd2009-12-21 16:28:53 -0800706 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
Nick Piggind208bbd2009-12-21 16:28:53 -0800707
Al Viro0aec09d2014-08-07 07:32:06 -0400708 if (remount_ro) {
709 if (sb->s_pins.first) {
710 up_write(&sb->s_umount);
Al Viro8fa1f1c2014-05-21 18:22:52 -0400711 sb_pin_kill(sb);
Al Viro0aec09d2014-08-07 07:32:06 -0400712 down_write(&sb->s_umount);
713 if (!sb->s_root)
714 return 0;
715 if (sb->s_writers.frozen != SB_UNFROZEN)
716 return -EBUSY;
717 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
718 }
719 }
720 shrink_dcache_sb(sb);
721
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 /* If we are remounting RDONLY and current sb is read/write,
723 make sure there are no rw files opened */
Nick Piggind208bbd2009-12-21 16:28:53 -0800724 if (remount_ro) {
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100725 if (force) {
Al Viroeee5cc22013-10-04 11:06:42 -0400726 sb->s_readonly_remount = 1;
727 smp_wmb();
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100728 } else {
729 retval = sb_prepare_remount_readonly(sb);
730 if (retval)
731 return retval;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 }
734
735 if (sb->s_op->remount_fs) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 retval = sb->s_op->remount_fs(sb, &flags, data);
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200737 if (retval) {
738 if (!force)
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100739 goto cancel_readonly;
Miklos Szeredi2833eb22011-10-28 14:13:20 +0200740 /* If forced remount, go ahead despite any errors */
741 WARN(1, "forced remount of a %s fs returned %i\n",
742 sb->s_type->name, retval);
743 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 }
745 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100746 /* Needs to be ordered wrt mnt_is_readonly() */
747 smp_wmb();
748 sb->s_readonly_remount = 0;
Christoph Hellwigc79d9672010-05-19 07:16:40 -0400749
Nick Piggind208bbd2009-12-21 16:28:53 -0800750 /*
751 * Some filesystems modify their metadata via some other path than the
752 * bdev buffer cache (eg. use a private mapping, or directories in
753 * pagecache, etc). Also file data modifications go via their own
754 * mappings. So If we try to mount readonly then copy the filesystem
755 * from bdev, we could get stale data, so invalidate it to give a best
756 * effort at coherency.
757 */
758 if (remount_ro && sb->s_bdev)
759 invalidate_bdev(sb->s_bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 return 0;
Miklos Szeredi4ed5e822011-11-21 12:11:31 +0100761
762cancel_readonly:
763 sb->s_readonly_remount = 0;
764 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765}
766
Jens Axboea2a95372009-03-17 09:38:40 +0100767static void do_emergency_remount(struct work_struct *work)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768{
Al Virodca33252010-07-25 02:31:46 +0400769 struct super_block *sb, *p = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
771 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400772 list_for_each_entry(sb, &super_blocks, s_list) {
Al Viroa5166162011-12-12 22:53:00 -0500773 if (hlist_unhashed(&sb->s_instances))
Al Viro551de6f2010-03-22 19:36:35 -0400774 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 sb->s_count++;
776 spin_unlock(&sb_lock);
Al Viro443b94b2009-05-05 23:48:50 -0400777 down_write(&sb->s_umount);
Al Virodabe0dc2012-01-03 21:01:29 -0500778 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
779 !(sb->s_flags & MS_RDONLY)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700780 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 * What lock protects sb->s_flags??
782 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 do_remount_sb(sb, MS_RDONLY, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 }
Al Viro443b94b2009-05-05 23:48:50 -0400785 up_write(&sb->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 spin_lock(&sb_lock);
Al Virodca33252010-07-25 02:31:46 +0400787 if (p)
788 __put_super(p);
789 p = sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 }
Al Virodca33252010-07-25 02:31:46 +0400791 if (p)
792 __put_super(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793 spin_unlock(&sb_lock);
Jens Axboea2a95372009-03-17 09:38:40 +0100794 kfree(work);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 printk("Emergency Remount complete\n");
796}
797
798void emergency_remount(void)
799{
Jens Axboea2a95372009-03-17 09:38:40 +0100800 struct work_struct *work;
801
802 work = kmalloc(sizeof(*work), GFP_ATOMIC);
803 if (work) {
804 INIT_WORK(work, do_emergency_remount);
805 schedule_work(work);
806 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700807}
808
809/*
810 * Unnamed block devices are dummy devices used by virtual
811 * filesystems which don't use real block-devices. -- jrs
812 */
813
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400814static DEFINE_IDA(unnamed_dev_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
Thomas Bächlera2a4dc42014-04-03 21:55:37 +0200816/* Many userspace utilities consider an FSID of 0 invalid.
817 * Always return at least 1 from get_anon_bdev.
818 */
819static int unnamed_dev_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820
Al Viro0ee5dc62011-07-07 15:44:25 -0400821int get_anon_bdev(dev_t *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700822{
823 int dev;
824 int error;
825
826 retry:
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400827 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700828 return -ENOMEM;
829 spin_lock(&unnamed_dev_lock);
Al Viroc63e09e2009-06-24 02:05:18 -0400830 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
Al Virof21f6222009-06-24 03:12:00 -0400831 if (!error)
832 unnamed_dev_start = dev + 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 spin_unlock(&unnamed_dev_lock);
834 if (error == -EAGAIN)
835 /* We raced and lost with another CPU. */
836 goto retry;
837 else if (error)
838 return -EAGAIN;
839
Tejun Heoe8c8d1b2013-02-27 17:05:04 -0800840 if (dev == (1 << MINORBITS)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 spin_lock(&unnamed_dev_lock);
Alexey Dobriyanad76cbc2008-08-28 06:26:23 +0400842 ida_remove(&unnamed_dev_ida, dev);
Al Virof21f6222009-06-24 03:12:00 -0400843 if (unnamed_dev_start > dev)
844 unnamed_dev_start = dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700845 spin_unlock(&unnamed_dev_lock);
846 return -EMFILE;
847 }
Al Viro0ee5dc62011-07-07 15:44:25 -0400848 *p = MKDEV(0, dev & MINORMASK);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849 return 0;
850}
Al Viro0ee5dc62011-07-07 15:44:25 -0400851EXPORT_SYMBOL(get_anon_bdev);
852
853void free_anon_bdev(dev_t dev)
854{
855 int slot = MINOR(dev);
856 spin_lock(&unnamed_dev_lock);
857 ida_remove(&unnamed_dev_ida, slot);
858 if (slot < unnamed_dev_start)
859 unnamed_dev_start = slot;
860 spin_unlock(&unnamed_dev_lock);
861}
862EXPORT_SYMBOL(free_anon_bdev);
863
864int set_anon_super(struct super_block *s, void *data)
865{
866 int error = get_anon_bdev(&s->s_dev);
867 if (!error)
868 s->s_bdi = &noop_backing_dev_info;
869 return error;
870}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871
872EXPORT_SYMBOL(set_anon_super);
873
874void kill_anon_super(struct super_block *sb)
875{
Al Viro0ee5dc62011-07-07 15:44:25 -0400876 dev_t dev = sb->s_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700877 generic_shutdown_super(sb);
Al Viro0ee5dc62011-07-07 15:44:25 -0400878 free_anon_bdev(dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879}
880
881EXPORT_SYMBOL(kill_anon_super);
882
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883void kill_litter_super(struct super_block *sb)
884{
885 if (sb->s_root)
886 d_genocide(sb->s_root);
887 kill_anon_super(sb);
888}
889
890EXPORT_SYMBOL(kill_litter_super);
891
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700892static int ns_test_super(struct super_block *sb, void *data)
893{
894 return sb->s_fs_info == data;
895}
896
897static int ns_set_super(struct super_block *sb, void *data)
898{
899 sb->s_fs_info = data;
900 return set_anon_super(sb, NULL);
901}
902
Al Viroceefda62010-07-26 13:16:50 +0400903struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
904 void *data, int (*fill_super)(struct super_block *, void *, int))
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700905{
906 struct super_block *sb;
907
David Howells9249e172012-06-25 12:55:37 +0100908 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700909 if (IS_ERR(sb))
Al Viroceefda62010-07-26 13:16:50 +0400910 return ERR_CAST(sb);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700911
912 if (!sb->s_root) {
913 int err;
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700914 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
915 if (err) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400916 deactivate_locked_super(sb);
Al Viroceefda62010-07-26 13:16:50 +0400917 return ERR_PTR(err);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700918 }
919
920 sb->s_flags |= MS_ACTIVE;
921 }
922
Al Viroceefda62010-07-26 13:16:50 +0400923 return dget(sb->s_root);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700924}
925
Al Viroceefda62010-07-26 13:16:50 +0400926EXPORT_SYMBOL(mount_ns);
Serge E. Hallyn909e6d92009-04-06 19:01:07 -0700927
David Howells93614012006-09-30 20:45:40 +0200928#ifdef CONFIG_BLOCK
Linus Torvalds1da177e2005-04-16 15:20:36 -0700929static int set_bdev_super(struct super_block *s, void *data)
930{
931 s->s_bdev = data;
932 s->s_dev = s->s_bdev->bd_dev;
Jens Axboe32a88aa2009-09-16 15:02:33 +0200933
934 /*
935 * We set the bdi here to the queue backing, file systems can
936 * overwrite this in ->fill_super()
937 */
938 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939 return 0;
940}
941
942static int test_bdev_super(struct super_block *s, void *data)
943{
944 return (void *)s->s_bdev == data;
945}
946
Al Viro152a0832010-07-25 00:46:55 +0400947struct dentry *mount_bdev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 int flags, const char *dev_name, void *data,
Al Viro152a0832010-07-25 00:46:55 +0400949 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950{
951 struct block_device *bdev;
952 struct super_block *s;
Tejun Heod4d77622010-11-13 11:55:18 +0100953 fmode_t mode = FMODE_READ | FMODE_EXCL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 int error = 0;
955
Al Viro30c40d22008-02-22 19:50:45 -0500956 if (!(flags & MS_RDONLY))
957 mode |= FMODE_WRITE;
958
Tejun Heod4d77622010-11-13 11:55:18 +0100959 bdev = blkdev_get_by_path(dev_name, mode, fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 if (IS_ERR(bdev))
Al Viro152a0832010-07-25 00:46:55 +0400961 return ERR_CAST(bdev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
963 /*
964 * once the super is inserted into the list by sget, s_umount
965 * will protect the lockfs code from trying to start a snapshot
966 * while we are mounting
967 */
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200968 mutex_lock(&bdev->bd_fsfreeze_mutex);
969 if (bdev->bd_fsfreeze_count > 0) {
970 mutex_unlock(&bdev->bd_fsfreeze_mutex);
971 error = -EBUSY;
972 goto error_bdev;
973 }
David Howells9249e172012-06-25 12:55:37 +0100974 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC,
975 bdev);
Christoph Hellwig4fadd7b2009-08-03 23:28:06 +0200976 mutex_unlock(&bdev->bd_fsfreeze_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 if (IS_ERR(s))
David Howells454e2392006-06-23 02:02:57 -0700978 goto error_s;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979
980 if (s->s_root) {
981 if ((flags ^ s->s_flags) & MS_RDONLY) {
Al Viro74dbbdd2009-05-06 01:07:50 -0400982 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -0700983 error = -EBUSY;
984 goto error_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985 }
David Howells454e2392006-06-23 02:02:57 -0700986
Tejun Heo4f331f02010-07-20 15:18:07 -0700987 /*
988 * s_umount nests inside bd_mutex during
Tejun Heoe525fd82010-11-13 11:55:17 +0100989 * __invalidate_device(). blkdev_put() acquires
990 * bd_mutex and can't be called under s_umount. Drop
991 * s_umount temporarily. This is safe as we're
992 * holding an active reference.
Tejun Heo4f331f02010-07-20 15:18:07 -0700993 */
994 up_write(&s->s_umount);
Tejun Heod4d77622010-11-13 11:55:18 +0100995 blkdev_put(bdev, mode);
Tejun Heo4f331f02010-07-20 15:18:07 -0700996 down_write(&s->s_umount);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997 } else {
998 char b[BDEVNAME_SIZE];
999
Al Viro30c40d22008-02-22 19:50:45 -05001000 s->s_mode = mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
Pekka Enberge78c9a02006-01-08 01:03:39 -08001002 sb_set_blocksize(s, block_size(bdev));
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001003 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001004 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001005 deactivate_locked_super(s);
David Howells454e2392006-06-23 02:02:57 -07001006 goto error;
Greg Kroah-Hartmanfa675762006-02-22 09:39:02 -08001007 }
David Howells454e2392006-06-23 02:02:57 -07001008
1009 s->s_flags |= MS_ACTIVE;
Theodore Ts'o87d8fe12009-01-03 09:47:09 -05001010 bdev->bd_super = s;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011 }
1012
Al Viro152a0832010-07-25 00:46:55 +04001013 return dget(s->s_root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014
David Howells454e2392006-06-23 02:02:57 -07001015error_s:
1016 error = PTR_ERR(s);
1017error_bdev:
Tejun Heod4d77622010-11-13 11:55:18 +01001018 blkdev_put(bdev, mode);
David Howells454e2392006-06-23 02:02:57 -07001019error:
Al Viro152a0832010-07-25 00:46:55 +04001020 return ERR_PTR(error);
1021}
1022EXPORT_SYMBOL(mount_bdev);
1023
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024void kill_block_super(struct super_block *sb)
1025{
1026 struct block_device *bdev = sb->s_bdev;
Al Viro30c40d22008-02-22 19:50:45 -05001027 fmode_t mode = sb->s_mode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028
H Hartley Sweetenddbaaf32009-04-29 20:14:57 -04001029 bdev->bd_super = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 generic_shutdown_super(sb);
1031 sync_blockdev(bdev);
Tejun Heod4d77622010-11-13 11:55:18 +01001032 WARN_ON_ONCE(!(mode & FMODE_EXCL));
Tejun Heoe525fd82010-11-13 11:55:17 +01001033 blkdev_put(bdev, mode | FMODE_EXCL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001034}
1035
1036EXPORT_SYMBOL(kill_block_super);
David Howells93614012006-09-30 20:45:40 +02001037#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038
Al Viro3c26ff62010-07-25 11:46:36 +04001039struct dentry *mount_nodev(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 int flags, void *data,
Al Viro3c26ff62010-07-25 11:46:36 +04001041 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042{
1043 int error;
David Howells9249e172012-06-25 12:55:37 +01001044 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
1046 if (IS_ERR(s))
Al Viro3c26ff62010-07-25 11:46:36 +04001047 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001049 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001051 deactivate_locked_super(s);
Al Viro3c26ff62010-07-25 11:46:36 +04001052 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 }
1054 s->s_flags |= MS_ACTIVE;
Al Viro3c26ff62010-07-25 11:46:36 +04001055 return dget(s->s_root);
1056}
1057EXPORT_SYMBOL(mount_nodev);
1058
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059static int compare_single(struct super_block *s, void *p)
1060{
1061 return 1;
1062}
1063
Al Virofc14f2f2010-07-25 01:48:30 +04001064struct dentry *mount_single(struct file_system_type *fs_type,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 int flags, void *data,
Al Virofc14f2f2010-07-25 01:48:30 +04001066 int (*fill_super)(struct super_block *, void *, int))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001067{
1068 struct super_block *s;
1069 int error;
1070
David Howells9249e172012-06-25 12:55:37 +01001071 s = sget(fs_type, compare_single, set_anon_super, flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001072 if (IS_ERR(s))
Al Virofc14f2f2010-07-25 01:48:30 +04001073 return ERR_CAST(s);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 if (!s->s_root) {
Theodore Ts'o9b04c992006-03-24 03:15:10 -08001075 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 if (error) {
Al Viro74dbbdd2009-05-06 01:07:50 -04001077 deactivate_locked_super(s);
Al Virofc14f2f2010-07-25 01:48:30 +04001078 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 }
1080 s->s_flags |= MS_ACTIVE;
Kay Sievers9329d1b2009-12-18 21:18:15 +01001081 } else {
1082 do_remount_sb(s, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083 }
Al Virofc14f2f2010-07-25 01:48:30 +04001084 return dget(s->s_root);
1085}
1086EXPORT_SYMBOL(mount_single);
1087
Al Viro9d412a42011-03-17 22:08:28 -04001088struct dentry *
1089mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090{
Al Viroc96e41e2010-07-25 00:17:56 +04001091 struct dentry *root;
Al Viro9d412a42011-03-17 22:08:28 -04001092 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093 char *secdata = NULL;
Al Viro9d412a42011-03-17 22:08:28 -04001094 int error = -ENOMEM;
Al Viro80893522010-02-05 09:30:46 -05001095
Eric Parise0007522008-03-05 10:31:54 -05001096 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001097 secdata = alloc_secdata();
David Howells454e2392006-06-23 02:02:57 -07001098 if (!secdata)
Al Viro9d412a42011-03-17 22:08:28 -04001099 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
Eric Parise0007522008-03-05 10:31:54 -05001101 error = security_sb_copy_data(data, secdata);
David Howells454e2392006-06-23 02:02:57 -07001102 if (error)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 goto out_free_secdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 }
1105
Al Viro1a102ff2011-03-16 09:07:58 -04001106 root = type->mount(type, flags, name, data);
1107 if (IS_ERR(root)) {
1108 error = PTR_ERR(root);
1109 goto out_free_secdata;
Al Viroc96e41e2010-07-25 00:17:56 +04001110 }
Al Viro9d412a42011-03-17 22:08:28 -04001111 sb = root->d_sb;
1112 BUG_ON(!sb);
1113 WARN_ON(!sb->s_bdi);
Linus Torvalds6c510382011-03-24 10:16:26 -07001114 WARN_ON(sb->s_bdi == &default_backing_dev_info);
Al Viro9d412a42011-03-17 22:08:28 -04001115 sb->s_flags |= MS_BORN;
David Howells454e2392006-06-23 02:02:57 -07001116
Al Viro9d412a42011-03-17 22:08:28 -04001117 error = security_sb_kern_mount(sb, flags, secdata);
Jörn Engel5129a462010-04-25 08:54:42 +02001118 if (error)
1119 goto out_sb;
David Howells454e2392006-06-23 02:02:57 -07001120
Jeff Layton42cb56a2009-09-18 13:05:53 -07001121 /*
1122 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1123 * but s_maxbytes was an unsigned long long for many releases. Throw
1124 * this warning for a little while to try and catch filesystems that
Jeff Layton4358b562011-03-29 09:33:31 -04001125 * violate this rule.
Jeff Layton42cb56a2009-09-18 13:05:53 -07001126 */
Al Viro9d412a42011-03-17 22:08:28 -04001127 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1128 "negative value (%lld)\n", type->name, sb->s_maxbytes);
Jeff Layton42cb56a2009-09-18 13:05:53 -07001129
Al Viro9d412a42011-03-17 22:08:28 -04001130 up_write(&sb->s_umount);
Gerald Schaefer8680e222005-06-21 17:15:16 -07001131 free_secdata(secdata);
Al Viro9d412a42011-03-17 22:08:28 -04001132 return root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001133out_sb:
Al Viro9d412a42011-03-17 22:08:28 -04001134 dput(root);
1135 deactivate_locked_super(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136out_free_secdata:
1137 free_secdata(secdata);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138out:
David Howells454e2392006-06-23 02:02:57 -07001139 return ERR_PTR(error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140}
1141
Jan Kara5accdf82012-06-12 16:20:34 +02001142/*
1143 * This is an internal function, please use sb_end_{write,pagefault,intwrite}
1144 * instead.
1145 */
1146void __sb_end_write(struct super_block *sb, int level)
1147{
1148 percpu_counter_dec(&sb->s_writers.counter[level-1]);
1149 /*
1150 * Make sure s_writers are updated before we wake up waiters in
1151 * freeze_super().
1152 */
1153 smp_mb();
1154 if (waitqueue_active(&sb->s_writers.wait))
1155 wake_up(&sb->s_writers.wait);
1156 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_);
1157}
1158EXPORT_SYMBOL(__sb_end_write);
1159
1160#ifdef CONFIG_LOCKDEP
1161/*
1162 * We want lockdep to tell us about possible deadlocks with freezing but
1163 * it's it bit tricky to properly instrument it. Getting a freeze protection
1164 * works as getting a read lock but there are subtle problems. XFS for example
1165 * gets freeze protection on internal level twice in some cases, which is OK
1166 * only because we already hold a freeze protection also on higher level. Due
1167 * to these cases we have to tell lockdep we are doing trylock when we
1168 * already hold a freeze protection for a higher freeze level.
1169 */
1170static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock,
1171 unsigned long ip)
1172{
1173 int i;
1174
1175 if (!trylock) {
1176 for (i = 0; i < level - 1; i++)
1177 if (lock_is_held(&sb->s_writers.lock_map[i])) {
1178 trylock = true;
1179 break;
1180 }
1181 }
1182 rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip);
1183}
1184#endif
1185
1186/*
1187 * This is an internal function, please use sb_start_{write,pagefault,intwrite}
1188 * instead.
1189 */
1190int __sb_start_write(struct super_block *sb, int level, bool wait)
1191{
1192retry:
1193 if (unlikely(sb->s_writers.frozen >= level)) {
1194 if (!wait)
1195 return 0;
1196 wait_event(sb->s_writers.wait_unfrozen,
1197 sb->s_writers.frozen < level);
1198 }
1199
1200#ifdef CONFIG_LOCKDEP
1201 acquire_freeze_lock(sb, level, !wait, _RET_IP_);
1202#endif
1203 percpu_counter_inc(&sb->s_writers.counter[level-1]);
1204 /*
1205 * Make sure counter is updated before we check for frozen.
1206 * freeze_super() first sets frozen and then checks the counter.
1207 */
1208 smp_mb();
1209 if (unlikely(sb->s_writers.frozen >= level)) {
1210 __sb_end_write(sb, level);
1211 goto retry;
1212 }
1213 return 1;
1214}
1215EXPORT_SYMBOL(__sb_start_write);
1216
1217/**
1218 * sb_wait_write - wait until all writers to given file system finish
1219 * @sb: the super for which we wait
1220 * @level: type of writers we wait for (normal vs page fault)
1221 *
1222 * This function waits until there are no writers of given type to given file
1223 * system. Caller of this function should make sure there can be no new writers
1224 * of type @level before calling this function. Otherwise this function can
1225 * livelock.
1226 */
1227static void sb_wait_write(struct super_block *sb, int level)
1228{
1229 s64 writers;
1230
1231 /*
1232 * We just cycle-through lockdep here so that it does not complain
1233 * about returning with lock to userspace
1234 */
1235 rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_);
1236 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_);
1237
1238 do {
1239 DEFINE_WAIT(wait);
1240
1241 /*
1242 * We use a barrier in prepare_to_wait() to separate setting
1243 * of frozen and checking of the counter
1244 */
1245 prepare_to_wait(&sb->s_writers.wait, &wait,
1246 TASK_UNINTERRUPTIBLE);
1247
1248 writers = percpu_counter_sum(&sb->s_writers.counter[level-1]);
1249 if (writers)
1250 schedule();
1251
1252 finish_wait(&sb->s_writers.wait, &wait);
1253 } while (writers);
1254}
1255
Josef Bacik18e9e512010-03-23 10:34:56 -04001256/**
Randy Dunlap7000d3c2010-05-24 22:22:34 -07001257 * freeze_super - lock the filesystem and force it into a consistent state
1258 * @sb: the super to lock
Josef Bacik18e9e512010-03-23 10:34:56 -04001259 *
1260 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1261 * freeze_fs. Subsequent calls to this without first thawing the fs will return
1262 * -EBUSY.
Jan Kara5accdf82012-06-12 16:20:34 +02001263 *
1264 * During this function, sb->s_writers.frozen goes through these values:
1265 *
1266 * SB_UNFROZEN: File system is normal, all writes progress as usual.
1267 *
1268 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New
1269 * writes should be blocked, though page faults are still allowed. We wait for
1270 * all writes to complete and then proceed to the next stage.
1271 *
1272 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked
1273 * but internal fs threads can still modify the filesystem (although they
1274 * should not dirty new pages or inodes), writeback can run etc. After waiting
1275 * for all running page faults we sync the filesystem which will clean all
1276 * dirty pages and inodes (no new dirty pages or inodes can be created when
1277 * sync is running).
1278 *
1279 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs
1280 * modification are blocked (e.g. XFS preallocation truncation on inode
1281 * reclaim). This is usually implemented by blocking new transactions for
1282 * filesystems that have them and need this additional guard. After all
1283 * internal writers are finished we call ->freeze_fs() to finish filesystem
1284 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is
1285 * mostly auxiliary for filesystems to verify they do not modify frozen fs.
1286 *
1287 * sb->s_writers.frozen is protected by sb->s_umount.
Josef Bacik18e9e512010-03-23 10:34:56 -04001288 */
1289int freeze_super(struct super_block *sb)
1290{
1291 int ret;
1292
1293 atomic_inc(&sb->s_active);
1294 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001295 if (sb->s_writers.frozen != SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001296 deactivate_locked_super(sb);
1297 return -EBUSY;
1298 }
1299
Al Virodabe0dc2012-01-03 21:01:29 -05001300 if (!(sb->s_flags & MS_BORN)) {
1301 up_write(&sb->s_umount);
1302 return 0; /* sic - it's "nothing to do" */
1303 }
1304
Josef Bacik18e9e512010-03-23 10:34:56 -04001305 if (sb->s_flags & MS_RDONLY) {
Jan Kara5accdf82012-06-12 16:20:34 +02001306 /* Nothing to do really... */
1307 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001308 up_write(&sb->s_umount);
1309 return 0;
1310 }
1311
Jan Kara5accdf82012-06-12 16:20:34 +02001312 /* From now on, no new normal writers can start */
1313 sb->s_writers.frozen = SB_FREEZE_WRITE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001314 smp_wmb();
1315
Jan Kara5accdf82012-06-12 16:20:34 +02001316 /* Release s_umount to preserve sb_start_write -> s_umount ordering */
1317 up_write(&sb->s_umount);
1318
1319 sb_wait_write(sb, SB_FREEZE_WRITE);
1320
1321 /* Now we go and block page faults... */
1322 down_write(&sb->s_umount);
1323 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT;
1324 smp_wmb();
1325
1326 sb_wait_write(sb, SB_FREEZE_PAGEFAULT);
1327
1328 /* All writers are done so after syncing there won't be dirty data */
Josef Bacik18e9e512010-03-23 10:34:56 -04001329 sync_filesystem(sb);
1330
Jan Kara5accdf82012-06-12 16:20:34 +02001331 /* Now wait for internal filesystem counter */
1332 sb->s_writers.frozen = SB_FREEZE_FS;
Josef Bacik18e9e512010-03-23 10:34:56 -04001333 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001334 sb_wait_write(sb, SB_FREEZE_FS);
Josef Bacik18e9e512010-03-23 10:34:56 -04001335
Josef Bacik18e9e512010-03-23 10:34:56 -04001336 if (sb->s_op->freeze_fs) {
1337 ret = sb->s_op->freeze_fs(sb);
1338 if (ret) {
1339 printk(KERN_ERR
1340 "VFS:Filesystem freeze failed\n");
Jan Kara5accdf82012-06-12 16:20:34 +02001341 sb->s_writers.frozen = SB_UNFROZEN;
Kazuya Mioe1616302011-12-01 16:51:07 +09001342 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001343 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001344 deactivate_locked_super(sb);
1345 return ret;
1346 }
1347 }
Jan Kara5accdf82012-06-12 16:20:34 +02001348 /*
1349 * This is just for debugging purposes so that fs can warn if it
1350 * sees write activity when frozen is set to SB_FREEZE_COMPLETE.
1351 */
1352 sb->s_writers.frozen = SB_FREEZE_COMPLETE;
Josef Bacik18e9e512010-03-23 10:34:56 -04001353 up_write(&sb->s_umount);
1354 return 0;
1355}
1356EXPORT_SYMBOL(freeze_super);
1357
1358/**
1359 * thaw_super -- unlock filesystem
1360 * @sb: the super to thaw
1361 *
1362 * Unlocks the filesystem and marks it writeable again after freeze_super().
1363 */
1364int thaw_super(struct super_block *sb)
1365{
1366 int error;
1367
1368 down_write(&sb->s_umount);
Jan Kara5accdf82012-06-12 16:20:34 +02001369 if (sb->s_writers.frozen == SB_UNFROZEN) {
Josef Bacik18e9e512010-03-23 10:34:56 -04001370 up_write(&sb->s_umount);
1371 return -EINVAL;
1372 }
1373
1374 if (sb->s_flags & MS_RDONLY)
1375 goto out;
1376
1377 if (sb->s_op->unfreeze_fs) {
1378 error = sb->s_op->unfreeze_fs(sb);
1379 if (error) {
1380 printk(KERN_ERR
1381 "VFS:Filesystem thaw failed\n");
Josef Bacik18e9e512010-03-23 10:34:56 -04001382 up_write(&sb->s_umount);
1383 return error;
1384 }
1385 }
1386
1387out:
Jan Kara5accdf82012-06-12 16:20:34 +02001388 sb->s_writers.frozen = SB_UNFROZEN;
Josef Bacik18e9e512010-03-23 10:34:56 -04001389 smp_wmb();
Jan Kara5accdf82012-06-12 16:20:34 +02001390 wake_up(&sb->s_writers.wait_unfrozen);
Josef Bacik18e9e512010-03-23 10:34:56 -04001391 deactivate_locked_super(sb);
1392
1393 return 0;
1394}
1395EXPORT_SYMBOL(thaw_super);